CN104881643B - A kind of quick remnant object detection method and system - Google Patents
A kind of quick remnant object detection method and system Download PDFInfo
- Publication number
- CN104881643B CN104881643B CN201510268000.4A CN201510268000A CN104881643B CN 104881643 B CN104881643 B CN 104881643B CN 201510268000 A CN201510268000 A CN 201510268000A CN 104881643 B CN104881643 B CN 104881643B
- Authority
- CN
- China
- Prior art keywords
- target
- filtering
- static target
- foreground
- obtains
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
Abstract
The invention discloses a kind of quick remnant object detection methods and system, method to include：Video frame is read, and foreground detection is carried out according to the video frame of reading, obtains foreground target；Connected area disposal$ is carried out to foreground target, obtains the size and location of foreground target；The filtration that the filtering of sliding average coefficient is filtered and weighted based on dispersion degree is used to carry out static target tracking to the foreground target after Connected area disposal$, obtains candidate static target；The cascade filtration structure based on similarity filtering, NCC filterings and the filtering of angle point number is used to carry out static target classification to candidate static target, obtains legacy；Legacy according to obtaining is alarmed.The present invention is comprehensive to employ the filtration that the filtering of sliding average coefficient is filtered and weighted based on dispersion degree and the cascade filtration structure based on similarity filtering, NCC filterings and the filtering of angle point number, wrong report phenomenon, can be widely applied to field of video monitoring caused by ghost, shadow and pedestrian can be effectively reduced.
Description
Technical field
The present invention relates to field of video monitoring, especially a kind of quick remnant object detection method and system.
Background technology
With the fast development of science and technology and being becoming better and approaching perfection day by day for information technology, people are right in order to improve the quality of living
High-tech convenience, safety, the requirement of high efficiency become higher and higher.After the generation of U.S.'s September 11 attacks, instead
Probably become the most important thing of every country trouble free service, strengthen the precautionary measures to insecurity, provide one to people
The living environment of safety, the safety paid much attention to from strike action of terror to the full extent into various countries worldwide are asked
Topic.Especially in the bigger place of movement of population amount, safety problem is even more the problem of can not be ignored.By taking airport as an example, in current
There is more than 160 airport currently in use in the country of state, and with the rapid development of China's economic, the flow of the people and cargo on airport
Upwards of movement gradually increases, the White Cloud Airport in Guangzhou, the Pekinese Capital Airport, Shanghai pudong airport the annual passenger traffic volume with thousand
Ten thousand meters, under this densely-populated environment, trouble free service just highlights extremely important.So for densely populated public affairs
Place and the higher unit of some security levels and department, which carry out real-time, round-the-clock video monitoring, altogether just seems especially heavy
It will.
Video monitoring system is obtained in all trades and professions and is widely applied.In actual life, residential area, supermarket, silver
There is video monitoring equipment in row, airport, subway, museum etc..Under normal circumstances, more than monitoring system is mainly closed by traditional
What road TV monitoring and controlling CCTV (Closed-Circuit Television) was formed, monitoring scene can be recorded and be deposited
The video of storage, record and storage is mainly used for retrospectant evidence presentation, criminal offences can not be waited to send out in time to endangering public security
Go out warning, and staff is needed to monitor monitored picture constantly.Since monitoring point is relatively more, and it is difficult to accomplish monitored picture
Full display, staff can feel unable to do what one wants very much to do to so many monitored picture.If by the study found that one
2 tunnels or multi-path monitoring video was observed continuously more than 22 minutes in people, then he will miss 95% monitoring scene, it is difficult to find different
Reason condition, but these monitoring informations missed are possible to just be very important information, it is also possible to just occur in this period
Many criminal activities for endangering social public security.Once there is criminal activity or the attack of terrorism, staff can only pass through
The monitor video of each record is inquired and checked to manual mode, to find time, place and personage of event generation etc.
Information.At this point, because criminal activity or the attack of terrorism cause the loss that can not be retrieved.Therefore the video prison based on CCTV
Control system can not meet the needs of people are to safety precaution, and it is so good that monitoring effect is also expected there is no people.
Legacy detection technique based on computer vision is the advanced processes link of object state analysis, is intelligent video
One of challenging forward position direction of scientific rersearch in the important component of monitoring and at present computer realm.It is regarded in safety
In frequency monitoring system, it is widely used to the detection for leaving object in many fields, such as：For solving real time monitoring ground
The problem of luggage that the public arenas such as iron, station, market and large-scale square are lost, parked vehicle.
Leave object and be removed object detection be many video monitoring systems common task, have become intelligence regard
An important component in frequency monitoring system.Such as the detection and the inspection of unserviced luggage in public places of parking violation
It surveys.Leave analyte detection refers to by whether there is legacy in monitoring device monitor area（Such as luggage, package, fragment etc.）
Or other are deliberately retained in the object in monitoring area（Such as dangerous explosive）, in the event of above-mentioned suspicious object,
Legacy detecting system can send out alarm in time, and mark the position where legacy automatically in video.Such as：When one
The object of predefined rule is violated in sensitizing range（Such as subway, railway station, airport monitoring area）When being inside detained long
Between（The time presets according to different application scenarios）Or police will be sent out when being more than prespecified time value
Report.It leaves object detection and has become one of most important task in intelligent video monitoring system.In July, 08, in China Kunming
The case of explosion occurred on certain bus is as caused by the explosion object being retained on bus.It is counted according to relevant departments, " base
The extreme armed member such as ground " tissue Iraq for Irapi civilians, Iraq official and U.S. army various explosion and attack
It hits and already leads to more than 4200 US soldiers and ten thousand Irapi civilians are dead about more than 60, these injures and deaths are also mainly by thing
Caused by the self-control explosive first placed.
However, legacy detection technique in the industry has the following problems at present：
（1）Easily occur " ghost " phenomenon due to object is removed in monitoring area or go out when insensitive to illumination variation
Existing " shadow " phenomenon causes to report the generation of phenomenon by mistake, it is relatively low to detect true property；
（2）It can not reduce and report phenomenon by mistake because caused by pedestrian occurs in monitoring area, accuracy is relatively low.
Invention content
In order to solve the above-mentioned technical problem, the purpose of the present invention is：There is provided a kind of can effectively reduce ghost, shadow and pedestrian
Caused wrong report phenomenon, remnant object detection method accurately and quickly.
It is another object of the present invention to：It is a kind of can effectively reducing the caused wrong report phenomenon of ghost, shadow and pedestrian to provide,
Legacy detecting system accurately and quickly.
The technical solution adopted by the present invention to solve the technical problems is：
A kind of quick remnant object detection method, including：
A, video frame is read, and foreground detection is carried out according to the video frame of reading, obtains foreground target；
B, Connected area disposal$ is carried out to foreground target, obtains the size and location of foreground target；
C, the foreground target after Connected area disposal$ is used and the filtering of sliding average coefficient is filtered and weighted based on dispersion degree
Filtration carries out static target tracking, obtains candidate static target；
D, the cascade filtration knot based on similarity filtering, NCC filterings and the filtering of angle point number is used to candidate static target
Structure carries out static target classification, obtains legacy；
E, it is alarmed according to obtained legacy.
Further, the step B, including：
B1, medium filtering and Morphological scale-space are carried out to foreground target；
B2, Connected area disposal$ is carried out to the foreground target after Morphological scale-space, obtains the size and location of Blob object blocks.
Further, the step C, including：
C1, dispersion degree filtering is carried out to the foreground target after Connected area disposal$；
C2, static target tracking is carried out to the foreground target after dispersion degree filtering, filters out preliminary candidate static target；
C3, the filtering of sliding average coefficient is weighted to preliminary candidate static target, obtains candidate static target.
Further, the step C1 is specially：
It calculates the dispersion degree of each Blob object blocks, and dispersion degree is more than the Maximum single layer distribution threshold value threshold value of setting and less than setting
The blob object blocks removal of fixed minimum dispersion degree threshold value；Wherein, the dispersion degree of Blob object blocks is equal to Blob object block perimeters
Square divided by Blob object blocks area.
Further, the step C2, including：
C21, the area for obtaining blob object blocks calculate blob object blocks area and tracking rectangle frame area ratio, then
Blob object block of the ratio less than 0.75 is removed；
C22, the tracking for static target establish target tracker, and set the matching of target tracker and Blob object blocks
Condition, the target tracker include match counter and lose counter, and the match counter is used to record target following
The number that device and Blob object blocks match, the loss counter for record target tracker not with Blob object blocks
The matching condition of the number mixed, the target tracker and Blob object blocks includes target area ratio and center variation ratio
Example；
Current tracker and each Blob object blocks are carried out object matching by C23, the matching condition according to setting, if
With success, then match counter is enabled to add one；If matching is unsuccessful, into next target tracker；If all be used
Tracker be all unable to successful match, then it is assumed that the static target is a fresh target, establishes a new tracker at this time, so
The data of Blob object blocks are saved into new tracker afterwards；
The match counter of still tracker currently in use is added one by C24, all trackers of scanning, and for matching into
The tracker of work(retains the gray average on tracking rectangle frame corresponding position, and is the loss of the tracker of no successful match
Counter adds 1；
C25, judgement are that the value for the value or match counter for losing counter is more than the frequency threshold value of setting, if losing
The value of counter is more than the frequency threshold value of setting, then it is assumed that target is lost, and at this time removes current tracker；If match counter
Value be more than setting frequency threshold value, then current tracker is preliminary candidate static target.
Further, the step C3, including：
C31, the mean value that preliminary candidate static target retains gray average vector is calculated；
C32, the mean value calculated is subtracted to each element in gray average vector, obtains mean value difference vector；
C33,1*3 element weighted mean filters are carried out to mean value difference vector, the weights of the weighted mean filter are equal to
The label of value difference vector divided by the element number of mean value difference vector；
C34, the result vector of weighted mean filter is averaged to obtain preliminary candidate static target index slide it is flat
Mean value, and the Blob object blocks that index sliding average is more than setting average threshold are filtered out, obtain candidate static target.
Further, the step D, including：
D1, similarity filtering is carried out to candidate static target；
D2, NCC filterings are carried out to the candidate static target after similarity filtering；
D3, angle point number filtering is carried out to the candidate static target after NCC filterings, obtains legacy.
Further, the step D2 is specially：
The NCC coefficients of the foreground and background of the candidate static target corresponding position after similarity filtering are calculated, are then filtered
Fall the Blob object blocks that NCC coefficients are more than setting NCC threshold values.
Further, the step D3, including：
The angle point number of each Blob object blocks in the candidate static target after NCC filterings is calculated, then filters out angle point
Number is more than maximum angular points threshold value and the Blob object blocks less than minimum angle point number threshold value, and is made with the Blob object blocks left
For legacy.
Another technical solution is used by the present invention solves its technical problem：
A kind of quick legacy detecting system, including：
Foreground detection module for reading video frame, and carries out foreground detection according to the video frame of reading, obtains prospect mesh
Mark；
Connected area disposal$ module for carrying out Connected area disposal$ to foreground target, obtains the size and location of foreground target；
Static target tracking module is filtered and is weighted based on dispersion degree for being used to the foreground target after Connected area disposal$
The filtration of sliding average coefficient filtering carries out static target tracking, obtains candidate static target；
Static target sort module, for being used to candidate static target based on similarity filtering, NCC filterings and angle point
The cascade filtration structure of number filtering carries out static target classification, obtains legacy；
Alarm module, for being alarmed according to obtained legacy；
The output terminal of the foreground detection module passes sequentially through Connected area disposal$ module, Connected area disposal$ module, static mesh
It marks tracking module and static target sort module and then is connect with the input terminal of alarm module.
The beneficial effects of the method for the present invention is：It employs and the filtering of sliding average coefficient is filtered and weighted based on dispersion degree
Filtration carries out static target tracking, and employs the series connection based on similarity filtering, NCC filterings and the filtering of angle point number
It filters structure and carries out static target classification, similarity filtering, NCC filterings and angle point number, which are crossed, can effectively reduce shadow and ghost causes
Wrong report phenomenon, accuracy is high；Dispersion degree filtering, the filtering of weighting sliding average coefficient, NCC filterings and the filtering of angle point number can have
Wrong report phenomenon caused by effect reduces pedestrian, it is further provided the accuracy of detection.Further, in including being carried out to foreground target
The step of value filtering and Morphological scale-space, the foreground target for making acquisition are apparent.
The advantageous effect of system of the present invention is：Static target tracking module is employed to be filtered and weighted based on dispersion degree and be slided
The filtration of dynamic mean coefficient filtering carries out static target tracking, and static target sort module is then employed to be spent based on similar
The cascade filtration structure of filter, NCC filterings and the filtering of angle point number carries out static target classification, similarity filtering, NCC filterings and angle
Point number crosses wrong report phenomenon caused by effectively reducing shadow and ghost, and accuracy is high；Dispersion degree filtering, weighting sliding average system
Wrong report phenomenon caused by number filtering, NCC filterings and the filtering of angle point number can effectively reduce pedestrian, it is further provided the standard of detection
True property.
Description of the drawings
The invention will be further described with reference to the accompanying drawings and examples.
Fig. 1 is a kind of overall flow figure of quick remnant object detection method of the present invention；
Fig. 2 is the flow chart of step B of the present invention；
Fig. 3 is the flow chart of step C of the present invention；
Fig. 4 is the flow chart of step C2 of the present invention；
Fig. 5 is the flow chart of step C3 of the present invention；
Fig. 6 is the flow chart of step D of the present invention；
Fig. 7 is a kind of overall structure block diagram of quick legacy detecting system of the present invention；
Fig. 8 is the overall flow figure of one legacy detection algorithm of the embodiment of the present invention；
Fig. 9 is the overall flow figure that one legacy of the embodiment of the present invention differentiates the stage；
Figure 10 is the schematic diagram of cascaded structure in the classification of one static target of the embodiment of the present invention.
Specific embodiment
Reference Fig. 1, a kind of quick remnant object detection method, including：
A, video frame is read, and foreground detection is carried out according to the video frame of reading, obtains foreground target；
B, Connected area disposal$ is carried out to foreground target, obtains the size and location of foreground target；
C, the foreground target after Connected area disposal$ is used and the filtering of sliding average coefficient is filtered and weighted based on dispersion degree
Filtration carries out static target tracking, obtains candidate static target；
D, the cascade filtration knot based on similarity filtering, NCC filterings and the filtering of angle point number is used to candidate static target
Structure carries out static target classification, obtains legacy；
E, it is alarmed according to obtained legacy.
With reference to Fig. 2, it is further used as preferred embodiment, the step B, including：
B1, medium filtering and Morphological scale-space are carried out to foreground target；
B2, Connected area disposal$ is carried out to the foreground target after Morphological scale-space, obtains the size and location of Blob object blocks.
With reference to Fig. 3, it is further used as preferred embodiment, the step C, including：
C1, dispersion degree filtering is carried out to the foreground target after Connected area disposal$；
C2, static target tracking is carried out to the foreground target after dispersion degree filtering, filters out preliminary candidate static target；
C3, the filtering of sliding average coefficient is weighted to preliminary candidate static target, obtains candidate static target.
Preferred embodiment is further used as, the step C1 is specially：
It calculates the dispersion degree of each Blob object blocks, and dispersion degree is more than the Maximum single layer distribution threshold value threshold value of setting and less than setting
The blob object blocks removal of fixed minimum dispersion degree threshold value；Wherein, the dispersion degree of Blob object blocks is equal to Blob object block perimeters
Square divided by Blob object blocks area.
With reference to Fig. 4, it is further used as preferred embodiment, the step C2, including：
C21, the area for obtaining blob object blocks calculate blob object blocks area and tracking rectangle frame area ratio, then
Blob object block of the ratio less than 0.75 is removed；
C22, the tracking for static target establish target tracker, and set the matching of target tracker and Blob object blocks
Condition, the target tracker include match counter and lose counter, and the match counter is used to record target following
The number that device and Blob object blocks match, the loss counter for record target tracker not with Blob object blocks
The matching condition of the number mixed, the target tracker and Blob object blocks includes target area ratio and center variation ratio
Example；
Current tracker and each Blob object blocks are carried out object matching by C23, the matching condition according to setting, if
With success, then match counter is enabled to add one；If matching is unsuccessful, into next target tracker；If all be used
Tracker be all unable to successful match, then it is assumed that the static target is a fresh target, establishes a new tracker at this time, so
The data of Blob object blocks are saved into new tracker afterwards；
The match counter of still tracker currently in use is added one by C24, all trackers of scanning, and for matching into
The tracker of work(retains the gray average on tracking rectangle frame corresponding position, and is the loss of the tracker of no successful match
Counter adds 1；
C25, judgement are that the value for the value or match counter for losing counter is more than the frequency threshold value of setting, if losing
The value of counter is more than the frequency threshold value of setting, then it is assumed that target is lost, and at this time removes current tracker；If match counter
Value be more than setting frequency threshold value, then current tracker is preliminary candidate static target.
With reference to Fig. 5, it is further used as preferred embodiment, the step C3, including：
C31, the mean value that preliminary candidate static target retains gray average vector is calculated；
C32, the mean value calculated is subtracted to each element in gray average vector, obtains mean value difference vector；
C33,1*3 element weighted mean filters are carried out to mean value difference vector, the weights of the weighted mean filter are equal to
The label of value difference vector divided by the element number of mean value difference vector；
C34, the result vector of weighted mean filter is averaged to obtain preliminary candidate static target index slide it is flat
Mean value, and the Blob object blocks that index sliding average is more than setting average threshold are filtered out, obtain candidate static target.
With reference to Fig. 6, it is further used as preferred embodiment, the step D, including：
D1, similarity filtering is carried out to candidate static target；
D2, NCC filterings are carried out to the candidate static target after similarity filtering；
D3, angle point number filtering is carried out to the candidate static target after NCC filterings, obtains legacy.
Preferred embodiment is further used as, the step D2 is specially：
The NCC coefficients of the foreground and background of the candidate static target corresponding position after similarity filtering are calculated, are then filtered
Fall the Blob object blocks that NCC coefficients are more than setting NCC threshold values.
It is further used as preferred embodiment, the step D3, including：
The angle point number of each Blob object blocks in the candidate static target after NCC filterings is calculated, then filters out angle point
Number is more than maximum angular points threshold value and the Blob object blocks less than minimum angle point number threshold value, and is made with the Blob object blocks left
For legacy.
Reference Fig. 7, a kind of quick legacy detecting system, including：
Foreground detection module for reading video frame, and carries out foreground detection according to the video frame of reading, obtains prospect mesh
Mark；
Connected area disposal$ module for carrying out Connected area disposal$ to foreground target, obtains the size and location of foreground target；
Static target tracking module is filtered and is weighted based on dispersion degree for being used to the foreground target after Connected area disposal$
The filtration of sliding average coefficient filtering carries out static target tracking, obtains candidate static target；
Static target sort module, for being used to candidate static target based on similarity filtering, NCC filterings and angle point
The cascade filtration structure of number filtering carries out static target classification, obtains legacy；
Alarm module, for being alarmed according to obtained legacy；
The output terminal of the foreground detection module passes sequentially through Connected area disposal$ module, Connected area disposal$ module, static mesh
It marks tracking module and static target sort module and then is connect with the input terminal of alarm module.
The present invention is described in further detail with specific embodiment with reference to the accompanying drawings of the specification.
Embodiment one
The present embodiment to it is involved in the present invention to correlation theory and the present invention realization process illustrate.
（One）Object detection algorithms are left based on tracking
Based on tracking leave object detecting method refer to first carrying out all moving objects into monitoring scene with
Track simultaneously records relevant information, using information such as the directions of motion, speed, track of moving object, by specifically leaving quality testing
Object detection task is left in method of determining and calculating completion.The tracking of feature based, the tracking based on kinetic characteristic, the tracking based on region and
Tracking based on profile etc. is more commonly used method for tracking target.However legacy detection algorithm of the tradition based on tracking exists
The shortcomings that apparent：When monitoring scene is very crowded, the legacy detection algorithm based on tracking will fail.
（Two）The realization process of the present invention
Legacy detection algorithm of the present invention belongs to stays object detecting method based on tracking.And the present invention is to leaving
On the basis of analyte detection is built upon to resting tracking and judging, so there are one important premise --- restings by the present invention
Detection.
If some object is still in monitoring scene always, but the object is removed since sometime, such
Object is referred to as removing object." ghost " phenomenon is known as by the afterimage phenomenon that moving object occurs in the scene.Shadow is to work as
Object enters in scene, and irregular shape block is formed in the scene according into shadow projection since light is reflected or reflected.
Either removal analyte detection still leaves analyte detection and is required for carrying out in preset region, this region is known as
Defence area.Correspondingly, legacy is defined as follows：If some object is not present in defence area before, but in subsequent video
Appear in defence area and it is independent it is static be more than setting time threshold value, such object is referred to as legacy.
As shown in figure 8, legacy detection algorithm of the present invention is divided into two stages：First stage is background dimension
The shield stage is that foreground and background is separated, also referred to as moving object detection, can be " ghost " and real in this stage
Foreground object distinguishes；Second stage is that legacy differentiates the stage, i.e., the foreground object detected from the first stage point
Class is divided into static prospect and sport foreground, then carries out legacy and partial silence object（Remove object）Differentiation, lost
Object prospect is stayed, and is classified as legacy target.Second stage includes two sub-stages again：1. static target tracks sub-stage,
The sub-stage 2. static target is classified.The detail flowchart of second stage is as shown in Figure 9.
The main realization process of the present invention is introduced below.
（One）Foreground detection
The present invention detects the movement that occurs in video or static using the adaptive foreground segmentation method based on pixel
Target.In order to obtain clearly foreground target, it is also necessary to carry out medium filtering and Morphological scale-space to result images.
In image processing field, Morphological scale-space mainly includes four kinds of operations：Expansion, burn into open operation and close behaviour
Make.Corrosion is commonly used to eliminate some unnecessary zonules and isolated point in binary image；Expansion is the inverse operation of corrosion,
It is to fill up hole, connect small gap that it, which is acted on,.It is a process for first corroding reflation to open operation；Closed operation is after first expanding
One process of corrosion.The square structure element of present invention selection 3*3 carries out opening operation processing to bianry image.
（Two）Connected area disposal$
The purpose of Connected area disposal$ of the present invention is to obtain the size and location of foreground target.
After Morphological scale-space is completed, most isolated noise point can be eliminated in image, and smaller gap also can
It is connected, small hole is also filled with, but still might have the inside that larger hole is present in target.Observation is a large amount of
Bianry image it is found that under normal circumstances, the point on target location can substantially be linked to be bulk zone than comparatively dense.Foundation
Features described above, the present invention calculate the area of each connected region, and to those areas using detection of connectivity and scale filter method
It is abandoned as noise in region less than threshold value.It can thus ensure to detected the integrality and standard of target to a certain extent
Exactness, moreover it is possible to remove partial noise.
The present invention is referred to as Blob to foreground target pixel by the object block obtained after Connected area disposal$.
（Three）Static target tracks
The present invention establishes a target tracker for the tracking of static target, and target tracker includes match counter and loses
Lose counter.Match counter is used to record the number that tracker and current Blob are matched（Tracker and current Blob matchings
Current goal is represented to be tracked）.Losing counter and being used to recording tracker does not have the number matched with current Blob（Tracking
Device and current Blob mismatches represent current goal and are not tracked）.
Static target of the present invention tracks process to every frame Blob of acquisition into line trace, determines whether static target.Together
When the present invention certain filtering, such as size and dispersion degree filtering are also carried out to Blob during static target tracking.
（Four）Static target is classified
Static target classification is in order to carry out classification judgement to candidate static target, to remove interfering object（Such as shadow
Or pedestrian etc.）.Three filter conditions difference that present invention classification judges：Similarity, NCC and angle point number.These three filterings
Condition forms cascaded structure, and structure chart is as shown in Figure 10.
1st, similarity calculation
Similarity calculation is for calculating with reference to the box on the corresponding position of background and present frame（Detect rectangle frame）Gray scale
The similarity degree of value, it can filter out wrong report caused by shadow variation.
2nd, NCC is normal cross-correlation, after calculating foreground and background corresponding position box normalization
Related coefficient, it can filter wrong report caused by ghost and pedestrian.
3rd, angle point calculates
Angle point filtering detailed process be：Corner Detection is carried out in Blob block of pixels position, calculates Harris angle point numbers,
Then angle point number is more than max-thresholds and the Blob removals less than minimum threshold（According to priori, angle point number is too small
Blob be usually shadow, the excessive Blob of angle point number is usually pedestrian）.
Embodiment two
With reference to Fig. 1-10, the second embodiment of the present invention：
A kind of specific implementation process of quick remnant object detection method of the present invention includes the following steps：
Step 1：Read video frame.
Step 2：Foreground detection is carried out, obtains foreground target.
Step 3：Carry out medium filtering and Morphological scale-space.
Step 4：Connected area disposal$ is carried out, obtains blob size and locations.
Step 5：Its dispersion degree is calculated each Blob, and dispersion degree is more than max-thresholds and less than minimum threshold
Blob is removed；Wherein, dispersion degree=blob perimeters square/blob areas, the destination scatter in this step corresponding diagram 9 spends filter
Process；
Step 6：The area of blob is obtained, calculates blob areas and tracking rectangle frame area ratio, ratio is less than 0.75
Blob removal.
Step 7：Tracker is matched with each Blob according to matching condition, successful match then enables of tracker
With counter plus one；It is unsuccessful then to enter next tracker；If all trackers currently in use are all unsuccessful, then it is assumed that
It is a fresh target, establishes a new tracker at this time, Blob data is saved into new tracker.Wherein, it matches
Condition includes target area ratio and center variation ratio.
Step 8：Scan entire tracker, the match counter of still tracker currently in use added one, pair and target
Matched tracker will then retain the gray average on its tracking rectangle frame corresponding position, and to the tracking of no successful match
Device then enables loss counter add 1.
Step 9：If losing counter is more than some threshold value, then it is assumed that target is lost, and at this time removes current tracker
It removes.
Step 10：If match counter is more than some threshold value, current tracker is candidate static target, needs to count at this time
Calculate the index sliding average retained per frame gray average（Exponential Running Average——ERA）, this step pair
Static target ERA filter process in Fig. 9 is answered, specific implementation step is as follows：
A. the mean value that current goal retains gray average vector is calculated；
B. vectorial mean value subtracts each element of gray average vector, obtains mean value difference vector；
C. 1*3 element weighted mean filters are carried out to mean value difference vector, weighted mean filter weights are vector labelling/vector
Element number；
D. filter result vector is averaged to obtain ERA values, and filters out the Blob ERA more than setting average threshold.
Wherein, 1*3 elements weighted mean filter refers to, continuous three are extracted successively using the template of 1*3 sizes to vector
A element asks for weighted mean, and the element value in three element centre positions is then replaced with mean value, and each template slides backward one
A element position, until vector terminates.Such as existing vectorial [1,2,3,4,5], 1*3 template extractions to pixel value are { 1,2,3 },
Hypothesis weights are all 2, then weighted average is（2*1+2*2+2*3）/ 3=4, at this time template center position is replaced with weighted average 4
The value put, i.e., the value of vectorial second position, then result become [Isosorbide-5-Nitrae, 3,4,5], and so on until terminating.
Step 11：The similarity of the foreground and background of resting corresponding position is calculated, filters the larger Blob of similarity,
Static target similarity filter process in this step corresponding diagram 9；
Step 12：The NCC coefficients of the foreground and background of resting corresponding position are calculated, filter the larger Blob of NCC, this
Static target NCC filter process in step corresponding diagram 9；
Step 13：Calculate the angle point number of Blob, the filtering big Blob of angle point number is quiet in this step corresponding diagram 9
Only Corner number filter process；
Step 14：Object as legacy or is removed using the Blob left, sends out alarm.
Compared with prior art, the present invention has the following advantages：
（1）Employ filtered and weighted based on dispersion degree the filtering of sliding average coefficient filtration carry out static target with
Track, and employ the cascade filtration structure based on similarity filtering, NCC filterings and the filtering of angle point number and carry out static target point
Class, similarity filtering, NCC filterings and angle point number cross wrong report phenomenon caused by effectively reducing shadow and ghost, and accuracy is high；
Dispersion degree filtering, the filtering of weighting sliding average coefficient, NCC filterings and the filtering of angle point number are reported by mistake caused by effectively reducing pedestrian
Phenomenon, it is further provided the accuracy of detection.
（2）The step of including carrying out medium filtering and Morphological scale-space to foreground target, the foreground target for making acquisition is more clear
It is clear.
It is that the preferable of the present invention is implemented to be illustrated, but the invention is not limited to the implementation above
Example, those skilled in the art can also make various equivalent variations under the premise of without prejudice to spirit of the invention or replace
It changes, these equivalent deformations or replacement are all contained in the application claim limited range.
Claims (8)
1. a kind of quick remnant object detection method, it is characterised in that：Including：
A, video frame is read, and foreground detection is carried out according to the video frame of reading, obtains foreground target；
B, Connected area disposal$ is carried out to foreground target, obtains the size and location of foreground target；
C, the filtering that the filtering of sliding average coefficient is filtered and weighted based on dispersion degree is used to the foreground target after Connected area disposal$
Structure carries out static target tracking, obtains candidate static target；
D, to candidate static target use based on similarity filtering, NCC filtering and angle point number filtering cascade filtration structure into
Row static target is classified, and obtains legacy；
E, it is alarmed according to obtained legacy；
The step B, including：
B1, medium filtering and Morphological scale-space are carried out to foreground target；
B2, Connected area disposal$ is carried out to the foreground target after Morphological scale-space, obtains the size and location of Blob object blocks；
The step C, including：
C1, dispersion degree filtering is carried out to the foreground target after Connected area disposal$；
C2, static target tracking is carried out to the foreground target after dispersion degree filtering, filters out preliminary candidate static target；
C3, the filtering of sliding average coefficient is weighted to preliminary candidate static target, obtains candidate static target.
2. a kind of quick remnant object detection method according to claim 1, it is characterised in that：The step C1, tool
Body is：
The dispersion degree of each Blob object blocks is calculated, and dispersion degree is more than the Maximum single layer distribution threshold value threshold value of setting and less than setting
The blob object blocks removal of minimum dispersion degree threshold value；Wherein, the dispersion degree of Blob object blocks is equal to the flat of Blob object block perimeters
Side divided by the area of Blob object blocks.
3. a kind of quick remnant object detection method according to claim 2, it is characterised in that：The step C2, packet
It includes：
C21, the area for obtaining blob object blocks calculate blob object blocks area and tracking rectangle frame area ratio, then will compare
Blob object block of the value less than 0.75 removes；
C22, the tracking for static target establish target tracker, and set the matching item of target tracker and Blob object blocks
Part, the target tracker include match counter and lose counter, and the match counter is used to record target tracker
The number matched with Blob object blocks, it is described loss counter for record target tracker not with Blob target Block- matchings
On number, the matching condition of the target tracker and Blob object blocks includes target area ratio and center variation ratio；
Current tracker and each Blob object blocks are carried out object matching by C23, the matching condition according to setting, if matching into
Work(then enables match counter add one；If matching is unsuccessful, into next target tracker；If it is all it is currently in use with
Track device is all unable to successful match, then it is assumed that the static target is a fresh target, establishes a new tracker at this time, then will
The data of Blob object blocks are saved into new tracker；
The match counter of still tracker currently in use is added one, and is successful match by C24, all trackers of scanning
Tracker retains the gray average on tracking rectangle frame corresponding position, and is the loss counting of the tracker of no successful match
Device adds 1；
C25, judgement are that the value for the value or match counter for losing counter is more than the frequency threshold value of setting, are counted if losing
The value of device is more than the frequency threshold value of setting, then it is assumed that target is lost, and at this time removes current tracker；If the value of match counter
More than the frequency threshold value of setting, then current tracker is preliminary candidate static target.
4. a kind of quick remnant object detection method according to claim 3, it is characterised in that：The step C3, packet
It includes：
C31, the mean value that preliminary candidate static target retains gray average vector is calculated；
C32, the mean value calculated is subtracted to each element in gray average vector, obtains mean value difference vector；
C33,1*3 element weighted mean filters are carried out to mean value difference vector, the weights of the weighted mean filter are equal to equal value difference
The label of vector divided by the element number of mean value difference vector；
C34, the result vector of weighted mean filter is averaged to obtain the index sliding average of preliminary candidate static target,
And the Blob object blocks that index sliding average is more than setting average threshold are filtered out, obtain candidate static target.
5. a kind of quick remnant object detection method according to claim 1, it is characterised in that：The step D, packet
It includes：
D1, similarity filtering is carried out to candidate static target；
D2, NCC filterings are carried out to the candidate static target after similarity filtering；
D3, angle point number filtering is carried out to the candidate static target after NCC filterings, obtains legacy.
6. a kind of quick remnant object detection method according to claim 5, it is characterised in that：The step D2, tool
Body is：
The NCC coefficients of the foreground and background of the candidate static target corresponding position after similarity filtering are calculated, then filter out NCC
Coefficient is more than the Blob object blocks of setting NCC threshold values.
7. a kind of quick remnant object detection method according to claim 5, it is characterised in that：The step D3, packet
It includes：
The angle point number of each Blob object blocks in the candidate static target after NCC filterings is calculated, then filters out angle point number
It counts threshold value and the Blob object blocks less than minimum angle point number threshold value more than maximum angular, and the Blob object blocks to leave are as something lost
Stay object.
8. a kind of quick legacy detecting system, it is characterised in that：Including：
Foreground detection module for reading video frame, and carries out foreground detection according to the video frame of reading, obtains foreground target；
Connected area disposal$ module for carrying out Connected area disposal$ to foreground target, obtains the size and location of foreground target；
Static target tracking module is slided for using to be filtered and weighted based on dispersion degree to the foreground target after Connected area disposal$
The filtration of mean coefficient filtering carries out static target tracking, obtains candidate static target；
Static target sort module, for being used to candidate static target based on similarity filtering, NCC filterings and angle point number mistake
The cascade filtration structure of filter carries out static target classification, obtains legacy；
Alarm module, for being alarmed according to obtained legacy；
The output terminal of the foreground detection module pass sequentially through Connected area disposal$ module, Connected area disposal$ module, static target with
It track module and static target sort module and then is connect with the input terminal of alarm module；
The Connected area disposal$ module performs following operation successively：
Medium filtering and Morphological scale-space are carried out to foreground target；
Connected area disposal$ is carried out to the foreground target after Morphological scale-space, obtains the size and location of Blob object blocks；
The static target tracking module performs following operation successively：
Dispersion degree filtering is carried out to the foreground target after Connected area disposal$；
Foreground target after being filtered to dispersion degree carries out static target tracking, filters out preliminary candidate static target；
The filtering of sliding average coefficient is weighted to preliminary candidate static target, obtains candidate static target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510268000.4A CN104881643B (en) | 2015-05-22 | 2015-05-22 | A kind of quick remnant object detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510268000.4A CN104881643B (en) | 2015-05-22 | 2015-05-22 | A kind of quick remnant object detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104881643A CN104881643A (en) | 2015-09-02 |
| CN104881643B true CN104881643B (en) | 2018-06-12 |
Family
ID=53949131
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510268000.4A Active CN104881643B (en) | 2015-05-22 | 2015-05-22 | A kind of quick remnant object detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104881643B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101841948B1 (en) * | 2015-10-02 | 2018-03-26 | 엘지전자 주식회사 | Apparatus, Method and Mobile Terminal for providing service to prevent lost article on inner Vehicle |
| JP6390860B2 (en) * | 2016-01-25 | 2018-09-19 | パナソニックＩｐマネジメント株式会社 | Left object monitoring device, left object monitoring system including the same, and left object monitoring method |
| CN105740814B (en) * | 2016-01-29 | 2018-10-26 | 重庆扬讯软件技术股份有限公司 | A method of determining solid waste dangerous waste storage configuration using video analysis |
| CN108241837B (en) * | 2016-12-23 | 2022-02-01 | 亿阳信通股份有限公司 | Method and device for detecting remnants |
| CN107918762B (en) * | 2017-10-24 | 2022-01-14 | 江西省高速公路投资集团有限责任公司 | Rapid detection system and method for road scattered objects |
| CN111832349A (en) | 2019-04-18 | 2020-10-27 | 富士通株式会社 | Method and device for identifying error detection of carry-over object and image processing equipment |
| CN111832470A (en) * | 2020-07-15 | 2020-10-27 | 中兴飞流信息科技有限公司 | Remnant detection method integrating multiple models |
| CN111709404B (en) * | 2020-08-05 | 2024-01-12 | 广东电网有限责任公司 | Machine room legacy identification method, system and equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010084902A1 (en) * | 2009-01-22 | 2010-07-29 | 株式会社日立国際電気 | Intrusion alarm video processing device |
| CN102314695A (en) * | 2011-08-23 | 2012-01-11 | 北京黄金视讯科技有限公司 | Abandoned object detection method based on computer vision |
| CN102722700A (en) * | 2012-05-17 | 2012-10-10 | 浙江工商大学 | Method and system for detecting abandoned object in video monitoring |
| CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
-
2015
- 2015-05-22 CN CN201510268000.4A patent/CN104881643B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010084902A1 (en) * | 2009-01-22 | 2010-07-29 | 株式会社日立国際電気 | Intrusion alarm video processing device |
| CN102314695A (en) * | 2011-08-23 | 2012-01-11 | 北京黄金视讯科技有限公司 | Abandoned object detection method based on computer vision |
| CN102722700A (en) * | 2012-05-17 | 2012-10-10 | 浙江工商大学 | Method and system for detecting abandoned object in video monitoring |
| CN104156942A (en) * | 2014-07-02 | 2014-11-19 | 华南理工大学 | Detection method for remnants in complex environment |
Non-Patent Citations (3)
| Title |
|---|
| 一种基于双背景模型的遗留物检测方法;范俊君，战荫伟;《计算机系统应用》;20120819;201-205 * |
| 智能视频分析技术在银行业务中的研究与应用;范俊君;《中国优秀学位论文全文数据库 信息科技辑》;20120915;第20页第1-3段,第24页第1-3段，第26页第1-3段，第28页第1-3段，第29页第1-3段，第31页1段 * |
| 智能视频监控中的遗留物检测技术;张具琴，王海洋，胡振;《计算机与现代化》;20141231;54-57 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104881643A (en) | 2015-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104881643B (en) | A kind of quick remnant object detection method and system | |
| CN107527009B (en) | Remnant detection method based on YOLO target detection | |
| CN109887281B (en) | Method and system for monitoring traffic incident | |
| CN111160125B (en) | Railway foreign matter intrusion detection method based on railway monitoring | |
| WO2018130016A1 (en) | Parking detection method and device based on monitoring video | |
| EP1974326B1 (en) | Video signal analysis | |
| CN106056079B (en) | A kind of occlusion detection method of image capture device and human face five-sense-organ | |
| Tian et al. | Real-time detection of abandoned and removed objects in complex environments | |
| CN111161312B (en) | Object trajectory tracking and identifying device and system based on computer vision | |
| Kamthe et al. | Suspicious activity recognition in video surveillance system | |
| CN109948455B (en) | Detection method and device for left-behind object | |
| CN103106766A (en) | Forest fire identification method and forest fire identification system | |
| CN101727672A (en) | Method for detecting, tracking and identifying object abandoning/stealing event | |
| CN110781853B (en) | Crowd abnormality detection method and related device | |
| Malhi et al. | Vision based intelligent traffic management system | |
| CN110544271B (en) | Parabolic motion detection method and related device | |
| Patil et al. | Suspicious movement detection and tracking based on color histogram | |
| CN114708537A (en) | Multi-view-angle-based system and method for analyzing abnormal behaviors of complex places | |
| CN104184997A (en) | Quick surveillance video playback technique | |
| CN107124577A (en) | A kind of real-time alarm system for guarding against theft based on moving object detection | |
| CN111695545A (en) | Single-lane reverse driving detection method based on multi-target tracking | |
| CN116993265A (en) | Intelligent warehouse safety management system based on Internet of things | |
| KR101407394B1 (en) | System for abandoned and stolen object detection | |
| Al Jarouf et al. | A hybrid method to detect and verify vehicle crash with haar-like features and svm over the web | |
| CN113516691A (en) | High-altitude parabolic detection system based on machine vision |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |